An Adaptive Kalman Filter with Inaccurate Noise Covariance Matrix and Abnormal Dynamic Model

This paper proposes an adaptive Kalman filter (AKF) that can track state changes and estimate noise covariance matrix (NCM) to address the poor performance issue of existing AKFs in handling abnormal dynamic models when the initial NCM is inaccurate. This algorithm establishes an adaptive correction model for innovation, which corrects innovation contaminated by abnormal dynamic model to resist the effect of abnormal innovation on NCM estimation. A more robust and accurate estimation framework is then provided by fusing fading Kalman filter (FKF) and autocovariance least squares (ALS) for systems affected by abnormal dynamic model and inaccurate NCM. Simulation and experimental results from measured data both demonstrate that the proposed algorithm can simultaneously resist the effects of abnormal dynamic models and imprecise NCM on the filter, exhibiting better performance than existing AKFs.